Google launches Gemini 3 Flash and brings it to the Gemini app

Google expanded its model family today with Gemini 3 Flash, a version designed to respond quickly and reduce usage costs without giving up advanced reasoning capabilities. The launch matters because of an unusual distribution decision: the model is beginning to replace Gemini 2.5 Flash as the default option in the Gemini app and Search’s AI Mode.

Google’s strategy is clear: make a model capable of solving complex tasks the one millions of people use for everyday queries. For developers, Gemini 3 Flash is already available in preview through the Gemini API in AI Studio, Vertex AI and Gemini Enterprise, as well as Gemini CLI, Android Studio and the Google Antigravity platform.

More reasoning without the wait

The Flash series began as Google’s fast, affordable alternative to its Pro models. With Gemini 3 Flash, the company says that gap is narrowing: the model retains the reasoning foundation of the Gemini 3 family and adjusts how long it thinks based on the difficulty of each request.

Google says Gemini 3 Flash uses an average of 30% fewer tokens than Gemini 2.5 Pro on typical traffic to complete everyday tasks with better performance. Tokens are the units a model uses to divide and process text, code, images and other data; using fewer generally translates into lower costs and faster responses.

In the tests reported by the company, it scores 90.4% on GPQA Diamond, an advanced-level scientific question benchmark, and 33.7% on Humanity’s Last Exam without external tools. It also scores 81.2% on MMMU Pro, a multimodal understanding benchmark that combines text and images.

These results help compare models, but they do not directly guarantee how every application will perform. An assistant for analyzing contracts, a coding agent that modifies code or a search tool using local information require different data, instructions and controls. The significant development is that Google wants to bring these capabilities to mass-market products rather than reserve them for a slower or more expensive model.

A price designed for high-volume applications

Gemini 3 Flash costs $0.50 per million input tokens and $3 per million output tokens. Audio input is priced at $1 per million tokens. Input is the material sent to the model; output is the response it generates.

Google says Flash outperforms Gemini 2.5 Pro in quality and is three times faster, based on measurements from Artificial Analysis. The difference in price and latency is especially important for products that make many calls to a model: in-app assistants, document analysis, customer-service tools or interfaces that generate code as users work.

On SWE-bench Verified, an evaluation that measures whether an agent can resolve real-world software issues, Gemini 3 Flash scores 78%. Google says that beats both the 2.5 generation and Gemini 3 Pro. A Flash model beating the Pro version on a coding test illustrates how performance no longer depends solely on using the largest model.

From the API to Search and Gemini

The rollout will take place in stages worldwide. In the Gemini app, Flash replaces 2.5 Flash as the default model, meaning free users will receive the new capabilities without manually selecting a specific version.

Google highlights uses such as interpreting videos and images, turning an audio recording into a personalized quiz or creating small applications from spoken instructions. In Search’s AI Mode, Gemini 3 Flash will have to combine the model’s reasoning with up-to-date information and web links to answer multi-step requests, such as planning a trip or studying a complex subject.

For businesses, its arrival in Vertex AI and Gemini Enterprise makes it possible to test the model in controlled environments. The next test will be less eye-catching than the benchmarks: determining whether its speed, cost and tool-use capabilities hold up when it is integrated into real-world workflows. Google has set a high bar by making Flash its default option; it will now have to show that the decision noticeably improves the everyday experience.